Ares CEO: AI Chip Depreciation Hard to Predict, Pivoting to Data Centers
nashnova research
Asset manager Ares Management is steering clear of GPU financing, channeling capital into data centers and power infrastructure instead — CEO Michael Arougheti says he cannot underwrite the depreciation curve on chips, a choice that exposes a deepening split inside the AI infrastructure financing market.
Why is Ares avoiding GPU financing?
CEO Michael Arougheti told analysts that data-center financing is "simpler to underwrite and more attractive on a relative-value basis" than GPU deals.
He was blunt: he does not know "how to underwrite the depreciation and obsolescence curve" on chips. This means → GPUs iterate so fast that a chip can lose most of its value shortly after purchase, making residual-asset risk nearly impossible to model.
In plain terms = a data center is a building — it lasts decades. A GPU is a phone — outdated in two or three years. Ares is betting on the building, not the phone.
Where does the $5 trillion in AI capex go?
Arougheti estimates total AI capital spending at roughly $5 trillion: about $2 trillion for physical data centers and supporting power infrastructure, and about $3 trillion for GPU procurement.
He sees large demand for data-center financing but relatively limited available capital — a supply-demand mismatch that means better deal terms for lenders.
Ares will stick, "at least for now," to the $2 trillion half it knows best. This reflects a disciplined approach: even amid an AI boom, the firm is staying within its circle of competence.
What is going wrong in GPU financing?
Bonds issued by GPU cloud-leasing firm CoreWeave fell this month to levels that typically signal financial distress; its credit-default swap — CDS, essentially the cost of insuring against default — surged in tandem.
Meanwhile, Broadcom's roughly $60 billion chip-debt financing and SpaceX's talks to raise around $40 billion for Nvidia chips show the market is massive — but GPU depreciation cycles run just five to six years.
This means → borrowers may still be repaying the loan long after the chips have lost their value — precisely the risk Ares refuses to take.
What does this split mean for investors?
Ares's choice highlights a growing divide inside AI infrastructure financing: data centers are favored for their more stable asset profile and clearer underwriting logic.
The central question for GPU financing is whether technology-depreciation risk can be adequately priced. If it is underpriced, lenders face significant losses.
Put simply = the market is drawing a line between "AI's foundation" and "AI's chips." The first looks like traditional real estate; the second is a bet that the next generation of technology won't arrive too soon. Whoever reads the depreciation curve right earns the better return.
市场有风险，内容仅供研究参考，不构成投资建议。